Rotation effects of objective functions in parallel distributed multiobjective fuzzy genetics-based machine learning

Fuzzy genetics-based machine learning (FGBML) is one of data mining techniques using evolutionary computation. It can obtain fuzzy rule-based classifiers that are accurate and linguistically interpretable for human users. However, there are two major problems. One is that it is impossible to design the best classifier with respect to both accuracy and interpretability due to their tradeoff. To solve this problem, we proposed multiobjective FGBML (MoFGBML) where an evolutionary multiobjective optimization algorithm is used to obtain a number of classifiers with different tradeoffs between accuracy and complexity. The other is the heavy computational load of FGBML for large data sets. In the previous study, we applied parallel distributed implementation to our MoFGBML to overcome this problem. We examined the effects of the parallel distributed implementation on the search ability. Although the computational time became much shorter, the number of the obtained non-dominated classifiers became small. As a result, accurate classifiers were not obtained for some data sets. In this paper, we propose a simple idea to bias the search direction of our MoFGBML. We rotate one or two objective functions. This rotation changes the dominance relation in multiobjective optimization. Through computational experiments, we examine the effects of the rotated objective functions on the search ability of our MoFGBML for large data sets.